def test_optstate_sumsq(self):
"""Test that optstate sumsq and sumsq are computed correctly."""
init_fn, update_fn, _ = make_training_metrics(
self.num_train_steps,
optstate_sumsq_fields=['nu'],
optstate_sum_fields=['nu'])
initial_metrics_state = init_fn(self.mock_params0)
self.assertTrue(
pytree_equal(initial_metrics_state['optstate_sumsq'],
{'nu': jnp.zeros(self.num_train_steps)}))
self.assertTrue(
pytree_equal(initial_metrics_state['optstate_sum'],
{'nu': jnp.zeros(self.num_train_steps)}))
updated_metrics_state = update_fn(initial_metrics_state, 0,
self.mock_cost0, self.mock_grad1,
self.mock_params0, self.mock_params1,
self.mock_optimizer_state0)
updated_metrics_state = update_fn(updated_metrics_state, 1,
self.mock_cost1, self.mock_grad2,
self.mock_params1, self.mock_params2,
self.mock_optimizer_state1)
self.assertEqual(updated_metrics_state['optstate_sumsq']['nu'][0],
total_tree_norm_sql2(self.mock_nu0))
self.assertEqual(updated_metrics_state['optstate_sumsq']['nu'][1],
total_tree_norm_sql2(self.mock_nu1))
self.assertEqual(updated_metrics_state['optstate_sum']['nu'][0],
total_tree_sum(self.mock_nu0))
self.assertEqual(updated_metrics_state['optstate_sum']['nu'][1],
total_tree_sum(self.mock_nu1))
def test_update_param_norms(self):
"""Ensure that we update param norms correctly."""
init_fn, update_fn, _ = make_training_metrics(self.num_train_steps,
enable_param_norms=True)
initial_metrics_state = init_fn(self.mock_params0)
updated_metrics_state = update_fn(initial_metrics_state, 0,
self.mock_cost0, self.mock_grad1,
self.mock_params0, self.mock_params1,
self.mock_optimizer_state0)
updated_metrics_state = update_fn(updated_metrics_state, 1,
self.mock_cost1, self.mock_grad2,
self.mock_params1, self.mock_params2,
self.mock_optimizer_state1)
self.assertTrue(
pytree_equal(
updated_metrics_state['param_norms'], {
'foo':
jnp.array([
jnp.linalg.norm(self.mock_params0['foo']),
jnp.linalg.norm(self.mock_params1['foo']), 0.0, 0.0,
0.0
]),
'bar': {
'baz':
jnp.array([
jnp.linalg.norm(self.mock_params0['bar']['baz']),
jnp.linalg.norm(self.mock_params1['bar']['baz']),
0.0, 0.0, 0.0
])
}
}))
def test_summarize(self):
"""Test the training metrics summarizer."""
_, _, summarize_fn = make_training_metrics(self.num_train_steps,
enable_train_cost=True,
enable_ema=True)
metrics_state = {
'train_cost': jnp.array([1.0, 0.5, 0.25, 0.0, 0.0]),
'param_norm': {
'foo': 7.0,
'bar': {
'baz': 2.0
}
},
'grad_ema': {
'foo': 1 * jnp.ones(5),
'bar': {
'baz': 2 * jnp.ones(10)
}
},
'grad_sq_ema': {
'foo': 2 * jnp.ones(5),
'bar': {
'baz': 6 * jnp.ones(10)
}
},
'update_ema': {
'foo': 2 * jnp.ones(5),
'bar': {
'baz': 1 * jnp.ones(10)
}
},
'update_sq_ema': {
'foo': 6 * jnp.ones(5),
'bar': {
'baz': 2 * jnp.ones(10)
}
},
}
tree_summary = summarize_fn(metrics_state)
self.assertTrue(
pytree_equal(
tree_summary, {
'param_norm': {
'/foo': 7.0,
'/bar/baz': 2.0
},
'grad_var': {
'/foo': 5 * (2 - 1**2),
'/bar/baz': 10 * (6 - 2**2)
},
'update_var': {
'/foo': 5 * (6 - 2**2),
'/bar/baz': 10 * (2 - 1**2)
},
'update_ratio': {
'/foo': 5 * (6 - 2**2) / 7.0,
'/bar/baz': 10 * (2 - 1**2) / 2.0
}
}))
def test_init(self):
"""Test the training metrics initializer."""
zeros_like_params = jax.tree_map(jnp.zeros_like, self.mock_params0)
zeros_scalar_like_params = jax.tree_map(lambda x: 0.0,
self.mock_params0)
zeros_timeseries = jnp.zeros(self.num_train_steps)
zeros_timeseries_like_params = jax.tree_map(
lambda x: jnp.zeros(self.num_train_steps), self.mock_params0)
# Test init with everything disabled.
init_fn, _, _ = make_training_metrics(self.num_train_steps)
initial_metrics_state = init_fn(self.mock_params0)
self.assertTrue(
pytree_equal({'param_norm': zeros_scalar_like_params},
initial_metrics_state))
# Test init with enable_ema = True and enable_train_cost=True.
init_fn, _, _ = make_training_metrics(self.num_train_steps,
enable_ema=True,
enable_train_cost=True,
enable_param_norms=True,
enable_gradient_norm=True,
enable_update_norm=True,
enable_update_norms=True)
initial_metrics_state = init_fn(self.mock_params0)
self.assertTrue(
pytree_equal(
initial_metrics_state, {
'train_cost': zeros_timeseries,
'param_norm': zeros_scalar_like_params,
'grad_ema': zeros_like_params,
'grad_sq_ema': zeros_like_params,
'update_ema': zeros_like_params,
'update_sq_ema': zeros_like_params,
'param_norms': zeros_timeseries_like_params,
'gradient_norm': zeros_timeseries,
'update_norm': zeros_timeseries,
'update_norms': zeros_timeseries_like_params
}))
def test_update_update_norms(self):
"""Ensure that we update gradient and update norms correctly."""
init_fn, update_fn, _ = make_training_metrics(
self.num_train_steps,
enable_gradient_norm=True,
enable_update_norm=True,
enable_update_norms=True)
initial_metrics_state = init_fn(self.mock_params0)
updated_metrics_state = update_fn(initial_metrics_state, 0,
self.mock_cost0, self.mock_grad1,
self.mock_params0, self.mock_params1,
self.mock_optimizer_state0)
updated_metrics_state = update_fn(updated_metrics_state, 1,
self.mock_cost1, self.mock_grad2,
self.mock_params1, self.mock_params2,
self.mock_optimizer_state1)
self.assertTrue(
pytree_equal(
updated_metrics_state['update_norms'], {
'foo':
jnp.array([
self.step_size *
jnp.linalg.norm(self.mock_grad1['foo']),
self.step_size *
jnp.linalg.norm(self.mock_grad2['foo']), 0.0, 0.0, 0.0
]),
'bar': {
'baz':
jnp.array([
self.step_size *
jnp.linalg.norm(self.mock_grad1['bar']['baz']),
self.step_size *
jnp.linalg.norm(self.mock_grad2['bar']['baz']),
0.0, 0.0, 0.0
])
}
}))
self.assertEqual(updated_metrics_state['update_norm'][0],
total_tree_norm_l2(self.mock_grad1))
self.assertEqual(updated_metrics_state['update_norm'][1],
total_tree_norm_l2(self.mock_grad2))
self.assertEqual(updated_metrics_state['update_norm'][0],
self.step_size * total_tree_norm_l2(self.mock_grad1))
self.assertEqual(updated_metrics_state['update_norm'][1],
self.step_size * total_tree_norm_l2(self.mock_grad2))
def test_train_cost(self):
"""Ensure that the train cost is logged correctly."""
init_fn, update_fn, _ = make_training_metrics(self.num_train_steps,
enable_train_cost=True)
initial_metrics_state = init_fn(self.mock_params0)
updated_metrics_state = update_fn(initial_metrics_state, 0,
self.mock_cost0, self.mock_grad1,
self.mock_params0, self.mock_params1,
self.mock_optimizer_state0)
updated_metrics_state = update_fn(updated_metrics_state, 1,
self.mock_cost1, self.mock_grad2,
self.mock_params1, self.mock_params2,
self.mock_optimizer_state1)
self.assertTrue(
pytree_equal(
updated_metrics_state['train_cost'],
jnp.array([self.mock_cost0, self.mock_cost1, 0.0, 0.0, 0.0])))
def test_update_grad_ema(self):
"""Ensure that the training metrics updater updates grad ema correctly."""
init_fn, update_fn, _ = make_training_metrics(self.num_train_steps,
enable_ema=True,
ema_beta=0.5)
initial_metrics_state = init_fn(self.mock_params0)
updated_metrics_state = update_fn(initial_metrics_state, 0,
self.mock_cost0, self.mock_grad1,
self.mock_params0, self.mock_params1,
self.mock_optimizer_state0)
updated_metrics_state = update_fn(updated_metrics_state, 1,
self.mock_cost1, self.mock_grad2,
self.mock_params1, self.mock_params2,
self.mock_optimizer_state1)
self.assertTrue(
pytree_equal(
updated_metrics_state['grad_ema'],
jax.tree_map(lambda x, y, z: 0.25 * x + 0.25 * y + 0.5 * z,
self.mock_zeros, self.mock_grad1,
self.mock_grad2)))
def train(train_dir,
model,
dataset_builder,
initializer,
num_train_steps,
hps,
rng,
eval_batch_size,
eval_num_batches,
eval_train_num_batches,
eval_frequency,
checkpoint_steps,
early_stopping_target_name=None,
early_stopping_target_value=None,
early_stopping_mode=None,
eval_steps=None,
metrics_logger=None,
init_logger=None,
training_metrics_config=None,
callback_configs=None,
external_checkpoint_path=None):
"""Main training loop.
Trains the given network on the specified dataset for the given number of
epochs. Saves the training curve in train_dir/r=3/results.tsv.
Args:
train_dir: (str) Path of the training directory.
model: (BaseModel) Model object to be trained.
dataset_builder: dataset builder returned by datasets.get_dataset.
initializer: Must have API as defined in initializers.py
num_train_steps: (int) Number of steps to train on.
hps: (tf.HParams) Model, initialization and training hparams.
rng: (jax.random.PRNGKey) Rng seed used in model initialization and data
shuffling.
eval_batch_size: the evaluation batch size. If None, use hps.batch_size.
eval_num_batches: (int) The number of batches used for evaluating on
validation and test sets. Set to None to evaluate on the whole train set.
eval_train_num_batches: (int) The number of batches for evaluating on train.
Set to None to evaluate on the whole training set.
eval_frequency: (int) Evaluate every k steps.
checkpoint_steps: List of integers indicating special steps to save
checkpoints at. These checkpoints do not get used for preemption recovery.
early_stopping_target_name: A string naming the metric to use to perform
early stopping. If this metric reaches the value
`early_stopping_target_value`, training will stop. Must include the
dataset split (ex: validation/error_rate).
early_stopping_target_value: A float indicating the value at which to stop
training.
early_stopping_mode: One of "above" or "below", indicates if we should stop
when the metric is above or below the threshold value. Example: if
"above", then training will stop when
`report[early_stopping_target_name] >= early_stopping_target_value`.
eval_steps: List of integers indicating which steps to perform evals. If
provided, eval_frequency will be ignored. Performing an eval implies
saving a checkpoint that will be used to resume training in the case of
preemption.
metrics_logger: Used to log all eval metrics during training. See
utils.MetricLogger for API definition.
init_logger: Used for black box initializers that have learning curves.
training_metrics_config: Dict specifying the configuration of the
training_metrics_grabber. Set to None to skip logging of advanced training
metrics.
callback_configs: List of configs specifying general callbacks to run
during the eval phase. Empty list means no callbacks are run. See
callbacks.py for details on what is expected in a config.
external_checkpoint_path: (str) If this argument is set, we will load the
optimizer_state, params, batch_stats, and training_metrics from the
checkpoint at this location.
Yields:
metrics: A dictionary of all eval metrics from the given epoch.
"""
# NOTE: the initialization RNG should *not* be per-host, as this will create
# different sets of weights per host. However, all other RNGs should be
# per-host.
# TODO(znado,gilmer,gdahl): implement replicating the same initialization
# across hosts.
rng, init_rng = jax.random.split(rng)
rng = jax.random.fold_in(rng, jax.process_index())
rng, data_rng = jax.random.split(rng)
# only used if checkpoints_steps is non-empty.
checkpoint_dir = os.path.join(train_dir, 'checkpoints')
# For logging / processing off the main thread
pool = multiprocessing.pool.ThreadPool()
if jax.process_index() == 0:
logging.info('Let the training begin!')
logging.info('Dataset input shape: %r', hps.input_shape)
logging.info('Hyperparameters: %s', hps)
if eval_batch_size is None:
eval_batch_size = hps.batch_size
if callback_configs is None:
callback_configs = []
# Maybe run the initializer.
unreplicated_params, unreplicated_batch_stats = init_utils.initialize(
model.flax_module, initializer, model.loss_fn,
hps.input_shape, hps.output_shape, hps, init_rng, init_logger,
model.get_fake_batch(hps))
if jax.process_index() == 0:
utils.log_pytree_shape_and_statistics(unreplicated_params)
logging.info('train_size: %d,', hps.train_size)
# Note that global_step refers to the number of gradients calculations, not
# the number of model updates. This means when using gradient accumulation,
# one must supply configs where the number of steps are in units of gradient
# calculations, not model updates, and in post processing one must divide
# global_step by grad_accum_step_multiplier to get the number of updates.
#
# If using gradient accumulation, stretch the learning rate schedule by the
# number of gradient calculations per weight update.
stretch_factor = 1
if hps.get('total_accumulated_batch_size') is not None:
stretch_factor = hps.total_accumulated_batch_size // hps.batch_size
lr_fn = schedules.get_schedule_fn(hps.lr_hparams,
num_train_steps,
stretch_factor=stretch_factor)
optimizer_init_fn, optimizer_update_fn = optimizers.get_optimizer(
hps, model)
unreplicated_optimizer_state = optimizer_init_fn(unreplicated_params)
(unreplicated_metrics_state, metrics_update_fn,
metrics_summary_fn) = None, None, None
if training_metrics_config is not None:
(metrics_init_fn, metrics_update_fn,
metrics_summary_fn) = make_training_metrics(num_train_steps,
**training_metrics_config)
unreplicated_metrics_state = metrics_init_fn(unreplicated_params)
(optimizer_state, params, batch_stats, metrics_state, global_step,
sum_train_cost, preemption_count,
is_restored) = checkpoint.replicate_and_maybe_restore_checkpoint(
unreplicated_optimizer_state,
unreplicated_params,
unreplicated_batch_stats,
unreplicated_metrics_state,
train_dir=train_dir,
external_checkpoint_path=external_checkpoint_path)
if is_restored:
preemption_count += 1
# Fold the restored step into the dataset RNG so that we will get a
# different shuffle each time we restore, so that we do not repeat a
# previous dataset ordering again after restoring. This is not the only
# difference in shuffling each pre-emption, because we often times reshuffle
# the input files each time in a non-deterministic manner.
#
# Note that if we are pre-empted more than once per epoch then we will
# retrain more on the beginning of the training split, because each time we
# restore we refill the shuffle buffer with the first `shuffle_buffer_size`
# elements from the training split to continue training.
#
# Also note that for evaluating on the training split, because we are
# reshuffling each time, we will get a new eval_train split each time we are
# pre-empted.
data_rng = jax.random.fold_in(data_rng, global_step)
assert hps.batch_size % (jax.device_count()) == 0
assert eval_batch_size % (jax.device_count()) == 0
dataset = dataset_builder(
data_rng,
hps.batch_size,
eval_batch_size=eval_batch_size,
hps=hps,
)
update_fn = functools.partial(update,
training_cost=model.training_cost,
grad_clip=hps.get('grad_clip'),
optimizer_update_fn=optimizer_update_fn,
metrics_update_fn=metrics_update_fn)
# in_axes = (
# optimizer_state = 0,
# params = 0,
# batch_stats = 0,
# metrics_state = 0,
# batch = 0,
# step = None,
# lr = None,
# rng = None,
# local_device_index = 0,
# running_train_cost = 0,
# training_cost,
# grad_clip,
# optimizer_update_fn,
# metrics_state_update_fn)
# Also, we can donate buffers for 'optimizer', 'batch_stats',
# 'batch' and 'training_metrics_state' for update's pmapped computation.
update_pmapped = jax.pmap(update_fn,
axis_name='batch',
in_axes=(0, 0, 0, 0, 0, None, None, None, 0, 0),
donate_argnums=(0, 1, 2, 8))
# During eval, we can donate the 'batch' buffer. We don't donate the
# 'params' and 'batch_stats' buffers as we don't re-assign those values in
# eval, we do that only in train.
evaluate_batch_pmapped = jax.pmap(model.evaluate_batch,
axis_name='batch',
donate_argnums=(2, ))
start_time = time.time()
start_step = global_step
prev_eval_step = start_step
def get_step_frequency(cur_step):
return float(cur_step - start_step) / (time.time() - start_time)
if jax.process_index() == 0:
trainer_utils.log_message('Starting training!', pool, xm_work_unit)
# Numpy array of range(0, local_device_count) to send to each device to be
# folded into the RNG inside each train step to get a unique per-device RNG.
local_device_indices = np.arange(jax.local_device_count())
# Start at the resumed step and continue until we have finished the number of
# training steps. If building a dataset iterator using a tf.data.Dataset, in
# the case of a batch size that does not evenly divide the training dataset
# size, if using `ds.batch(..., drop_remainer=True)` on the training dataset
# then the final batch in this iterator will be a partial batch. However, if
# `drop_remainer=False`, then this iterator will always return batches of the
# same size, and the final batch will have elements from the start of the
# (num_epochs + 1)-th epoch.
train_iter = itertools.islice(dataset.train_iterator_fn(), global_step,
num_train_steps)
eval_callbacks = []
rng, callback_rng = jax.random.split(rng)
callback_rngs = jax.random.split(callback_rng, len(callback_configs))
for callback_rng, config in zip(callback_rngs, callback_configs):
eval_callback = callbacks.get_callback(config['callback_name'])(
model, params, batch_stats, optimizer_state, dataset, hps, config,
train_dir, callback_rng)
eval_callbacks.append(eval_callback)
train_iter = trainer_utils.prefetch_input_pipeline(
train_iter, hps.num_device_prefetches)
eval_start_time = start_time
eval_start_step = start_step
for _ in range(start_step, num_train_steps):
with jax.profiler.StepTraceAnnotation('train', step_num=global_step):
# NOTE(dsuo): to properly profile each step, we must include batch
# creation in the StepTraceContext (as opposed to putting `train_iter`
# directly in the top-level for loop).
batch = next(train_iter)
if global_step in checkpoint_steps and jax.process_index() == 0:
checkpoint.save_unreplicated_checkpoint_background(
checkpoint_dir,
optimizer_state,
params,
batch_stats,
metrics_state,
global_step,
preemption_count,
sum_train_cost,
max_to_keep=None)
lr = lr_fn(global_step)
optimizer_state, params, batch_stats, sum_train_cost, metrics_state, grad_norm = update_pmapped(
optimizer_state, params, batch_stats, metrics_state, batch,
global_step, lr, rng, local_device_indices, sum_train_cost)
global_step += 1
# TODO(gdahl, gilmer): consider moving this test up.
# NB: Since this test is after we increment global_step, having 0 in
# eval_steps does nothing.
if trainer_utils.should_eval(global_step, eval_frequency,
eval_steps):
train_steps_per_sec = (global_step - eval_start_step) / (
time.time() - eval_start_time)
eval_start_step = global_step
eval_start_time = time.time()
batch_stats = trainer_utils.maybe_sync_batchnorm_stats(
batch_stats)
report, eval_time = eval_metrics(params, batch_stats, dataset,
eval_num_batches,
eval_train_num_batches,
evaluate_batch_pmapped)
mean_train_cost = sum_train_cost.mean().item() / max(
1, global_step - prev_eval_step)
report.update(
learning_rate=float(lr),
global_step=global_step,
epoch=global_step * hps.batch_size // hps.train_size,
train_steps_per_sec=train_steps_per_sec,
overall_steps_per_sec=get_step_frequency(global_step),
eval_time=eval_time,
grad_norm=np.mean(grad_norm),
preemption_count=preemption_count,
train_cost=mean_train_cost)
for eval_callback in eval_callbacks:
callback_metrics = eval_callback.run_eval(
params, batch_stats, optimizer_state, global_step)
if set(callback_metrics.keys()).intersection(
set(report.keys())):
raise ValueError(
'There was a collision between the callback'
'metrics and the standard eval metrics keys')
report.update(callback_metrics)
yield report
if jax.process_index() == 0:
trainer_utils.log_eta(pool, xm_work_unit, global_step,
train_steps_per_sec, num_train_steps,
start_time, eval_frequency,
eval_steps, eval_time)
trainer_utils.log_epoch_report(report, metrics_logger)
trainer_utils.maybe_log_training_metrics(
metrics_state, metrics_summary_fn, metrics_logger)
checkpoint.save_unreplicated_checkpoint_background(
train_dir, optimizer_state, params, batch_stats,
metrics_state, global_step, preemption_count,
sum_train_cost)
sum_train_cost = jnp.zeros(jax.local_device_count())
prev_eval_step = global_step
early_stopping_condition = trainer_utils.check_for_early_stopping(
early_stopping_target_name, early_stopping_target_value,
early_stopping_mode, report)
if early_stopping_condition:
comparison_string = '>=' if early_stopping_mode == 'above' else '<='
logging.info(
'Early stopping because metric %s=%f, reached the target value '
'of %s %f.', early_stopping_target_name,
report[early_stopping_target_name], comparison_string,
early_stopping_target_value)
return
# Always log and checkpoint on host 0 at the end of training.
# If we moved where in the loop body evals happen then we would not need this
# test.
if prev_eval_step != num_train_steps:
train_steps_per_sec = (global_step - eval_start_step) / (
time.time() - eval_start_time)
batch_stats = trainer_utils.maybe_sync_batchnorm_stats(batch_stats)
report, eval_time = eval_metrics(params, batch_stats, dataset,
eval_num_batches,
eval_train_num_batches,
evaluate_batch_pmapped)
lr = lr_fn(global_step)
# Correct the average for the final partial epoch.
mean_train_cost = sum_train_cost.mean().item() / max(
1, global_step - prev_eval_step)
report.update(learning_rate=float(lr),
global_step=global_step,
epoch=global_step * hps.batch_size // hps.train_size,
train_steps_per_sec=train_steps_per_sec,
overall_steps_per_sec=get_step_frequency(global_step),
eval_time=eval_time,
grad_norm=np.mean(grad_norm),
preemption_count=preemption_count,
train_cost=mean_train_cost)
yield report
if jax.process_index() == 0:
trainer_utils.log_eta(pool, xm_work_unit, global_step,
train_steps_per_sec, num_train_steps,
start_time, eval_frequency, eval_steps,
eval_time)
trainer_utils.log_epoch_report(report, metrics_logger)
trainer_utils.maybe_log_training_metrics(metrics_state,
metrics_summary_fn,
metrics_logger)
checkpoint.save_unreplicated_checkpoint_background(
train_dir, optimizer_state, params, batch_stats, metrics_state,
global_step, preemption_count, sum_train_cost)
# To make sure the last checkpoint was correctly saved.
checkpoint.wait_for_checkpoint_save()